Question answering method and device for rail transit table-text mixed data

Abstract

The invention provides a question answering method and device for rail transit table-text mixed data, and relates to the technical field of artificial intelligence, and the method comprises the steps: inputting the rail transit table-text mixed data and questions into a question answering model, reasoning the question answering model according to a task instruction, and outputting a question answering result; wherein the task instruction is constructed based on a cognitive chain reasoning framework comprising five steps of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the task instruction comprises five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis. According to the method, the question and answer interpretability of the rail transit table-text mixed data is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a rail transit table-text hybrid data question and answer method and device. BACKGROUND

[0002] In the process of urban rail transit intelligence, a large amount of operation data is continuously generated in the form of coexistence of structured tables (such as train timetable, equipment parameter table, passenger flow statistics table) and unstructured text (such as dispatch log, fault report, operation specification).

[0003] The existing end-to-end table-text question and answer technology based on large language model relies on the zero-shot reasoning ability of general large language model, has certain generalization, but the existing model directly maps the question to the answer, lacks explicit modeling of the intermediate reasoning steps, causes logical discontinuity and error propagation, has the problem of uncontrollable reasoning process, and is difficult to meet the demand for explainability in the rail transit scene. SUMMARY

[0004] The present application provides a rail transit table-text hybrid data question and answer method and device to solve the defect of unexplainability in the existing rail transit table-text hybrid data question and answer, and to realize the improvement of the explainability of the rail transit table-text hybrid data question and answer.

[0005] The present application provides a rail transit table-text hybrid data question and answer method, comprising: inputting the rail transit table-text hybrid data and the question into a question and answer model, the question and answer model performing reasoning according to a task instruction, and outputting a question and answer result; The task instruction is constructed based on a cognitive chain reasoning framework including five steps of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the task instruction includes five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis.

[0006] In some embodiments, the inputting of the rail transit table-text hybrid data and the question into the question and answer model, the question and answer model performing reasoning according to the task instruction, and the outputting of the question and answer result, comprises: According to the sub-task instruction of information extraction, the question and answer model extracts key information related to the question from the rail transit table-text hybrid data; According to the sub-task instruction of logic construction, the question and answer model constructs reasoning logic based on the key information and professional knowledge and conventional rules in the rail transit field; According to the sub-task instruction of preliminary operation, the question and answer model performs operation on the key information according to the reasoning logic to obtain a preliminary answer; According to the sub-task instruction for checking the result, the question and answer model checks the preliminary answer to obtain a checking result; According to the sub-task instruction for analyzing the answer, the question and answer model optimizes and analyzes the preliminary answer according to the checking result, and outputs the question and answer result.

[0007] In some embodiments, the method further comprises: constructing a training sample containing the task instruction, table, text, question and answer; pre-training the question and answer model based on the training sample.

[0008] In some embodiments, the method further comprises: fine-tuning the pre-trained question and answer model using a low-rank adaptive matrix.

[0009] In some embodiments, the question and answer result includes the key information, the reasoning logic, the preliminary answer, the checking result, the final answer and the answer analysis.

[0010] In some embodiments, the question and answer model is a DeepSeek R1 large model.

[0011] The application also provides a question and answer device for rail transit table-text hybrid data, comprising: a cognitive chain reasoning module for inputting rail transit table-text hybrid data and a question into a question and answer model, wherein the question and answer model performs reasoning according to a task instruction and outputs a question and answer result; wherein the task instruction is constructed based on a cognitive chain reasoning framework including five steps of information extraction, logic construction, preliminary operation, result checking and answer analysis, and the task instruction includes five sub-task instructions of information extraction, logic construction, preliminary operation, result checking and answer analysis.

[0012] The application also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the question and answer method for rail transit table-text hybrid data according to any one of the above.

[0013] The application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the question and answer method for rail transit table-text hybrid data according to any one of the above.

[0014] The application also provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the question and answer method for rail transit table-text hybrid data according to any one of the above.

[0015] The application provides a table-text hybrid data question and answer method and device for rail transit, which inputs the table-text hybrid data for rail transit and a question into a question and answer model, and the question and answer model performs chain reasoning according to five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis in sequence, realizes explicit tracking of reasoning paths, and improves the question and answer explainability for the table-text hybrid data for rail transit, and breaks through the bottleneck of traditional end-to-end models in cross-modal data fusion and explainability. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is one of the flowcharts of the question and answer method for the table-text hybrid data for rail transit provided by the application.

[0018] Figure 2 is the second flowchart of the question and answer method for the table-text hybrid data for rail transit provided by the application.

[0019] Figure 3 is the structural schematic diagram of the question and answer device for the table-text hybrid data for rail transit provided by the application.

[0020] Figure 4 is the structural schematic diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the application more clear, the technical solutions in the application will be described clearly and completely in combination with the drawings in the application. Obviously, the described embodiments are some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0022] Figure 1 is one of the flowcharts of the question and answer method for the table-text hybrid data for rail transit provided by the application, as shown in Figure 1 The application provides a question and answer method for table-text hybrid data for rail transit, including the following steps: Step 110, input the rail transit table-text mixed data and the question into the question and answer model, and the question and answer model performs reasoning according to the task instruction and outputs a question and answer result. The task instruction is constructed based on a cognitive chain reasoning framework including five steps of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the task instruction includes five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis.

[0023] Specifically, the cognitive chain reasoning framework (CCRF) is a reasoning method simulating the human thinking process, which divides complex problems into multiple steps and structured cognitive chains to gradually advance reasoning to achieve more accurate and interpretable conclusions.

[0024] The cognitive chain reasoning framework constructs a progressive and tightly coupled reasoning link, and includes five indispensable core steps: information extraction, logic construction, preliminary operation, result verification and answer analysis.

[0025] In the processing of the rail transit field mixed data question and answer task integrating tables and texts, the task instruction is constructed based on the cognitive chain reasoning framework, the constructed task instruction includes the sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the complex question and answer task is divided into the chain reasoning process of information extraction, logic construction, preliminary operation, result verification and answer analysis through the five sub-task instructions.

[0026] The rail transit table-text mixed data and the question are input into the question and answer model, the question and answer model performs chain reasoning according to the five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis in turn, and outputs a question and answer result, thereby breaking through the bottleneck of the traditional end-to-end model in cross-modal data fusion and interpretability.

[0027] The question and answer method of the rail transit table-text mixed data provided by the application inputs the rail transit table-text mixed data and the question into the question and answer model, the question and answer model performs chain reasoning according to the five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis in turn, realizes explicit tracking of the reasoning path, thereby improving the question and answer interpretability for the rail transit table-text mixed data, and breaking through the bottleneck of the traditional end-to-end model in cross-modal data fusion and interpretability.

[0028] In some embodiments, the question and answer model is a DeepSeek R1 large model.

[0029] Specifically, the DeepSeek R1 large model is trained based on 100 billion parameters, can accurately understand complex problems including polysemous words, professional terms, context association, etc., generate fluent and accurate answers, support long text understanding (context window up to 128K), and is suitable for processing question and answer scenarios that require deep analysis. Therefore, the DeepSeek R1 large model is used as the question and answer model.

[0030] In some embodiments, Figure 2 is a flowchart of the question and answer method of the rail transit table-text hybrid data provided by the present application, as Figure 2 shown, the rail transit table-text hybrid data and the question are input into the question and answer model, the question and answer model performs reasoning according to the task instructions, and outputs the question and answer result, including: Step 210, according to the sub-task instruction of information extraction, the question and answer model extracts the key information related to the question from the rail transit table-text hybrid data.

[0031] Specifically, according to the requirement of the sub-task instruction of information extraction, the question and answer model extracts the key information related to the question from the rail transit table-text hybrid data. For example, the train number, station, time and other information in the train timetable, and the description of the line and operation situation in the text. After these information are extracted, they are stored in the specified variable according to the specific format (such as separating multiple value information with “#”), providing basic data for subsequent reasoning.

[0032] For example, the sub-task instruction of information extraction is to accurately extract various numerical and text information related to the question from the provided table and text, such as train number, station name, passenger flow, running time, etc. Store these information in the variable “{evidence}”, and separate multiple information with “#” symbol.

[0033] Step 220, according to the sub-task instruction of logic construction, the question and answer model constructs reasoning logic based on the key information and the professional knowledge and common rules in the rail transit field.

[0034] Specifically, according to the requirement of the sub-task instruction of logic construction, the question and answer model constructs the reasoning logic between the key information based on the extracted key information and the professional knowledge and common rules in the rail transit field. The reasoning logic can be a calculation formula or a logical judgment basis. The professional knowledge and common rules in the rail transit field can be embedded in the question and answer model in advance.

[0035] For example, when answering the question about passenger flow calculation, the question and answer model will generate the corresponding calculation formula or logical judgment basis according to the extracted passenger flow data of each station at different times, combined with the passenger flow calculation rules, and record it in the specified variable, clearly showing the operation relationship between the data.

[0036] For example, the subtask instruction of logical construction is: according to the extracted key information, combining the professional knowledge and rules in the field of rail transit, generating a calculation formula or logical judgment basis for answering the question, and storing it in the variable "{equation_or_logic}".

[0037] In step 230, according to the subtask instruction of preliminary operation, the question and answer model performs operation on the key information according to the reasoning logic to obtain a preliminary answer.

[0038] Specifically, according to the requirement of the subtask instruction of preliminary operation, the question and answer model performs calculation or reasoning on the key data according to the reasoning logic to obtain a preliminary answer. In the calculation process, the data consistency principle is strictly followed and the data unit is unified. The preliminary calculation or reasoning result is stored in the specified variable, and the key intermediate steps are recorded for subsequent checking and verification.

[0039] For example, the subtask instruction of preliminary operation is: according to the generated calculation formula or logical judgment basis, performing calculation or reasoning on the extracted key information to obtain a preliminary answer, and storing the preliminary answer in the variable "{preliminary_answer}".

[0040] In step 240, according to the subtask instruction of result verification, the question and answer model verifies the preliminary answer to obtain a verification result.

[0041] Specifically, according to the requirement of the subtask instruction of result verification, the question and answer model verifies the preliminary answer to obtain a verification result. The verification result includes error conditions and error correction methods.

[0042] The question and answer model will re-examine whether the calculation formula is correctly applied, whether the data reference is accurate, and whether the data extraction is complete. If errors are found, the error details (such as error type, occurrence position) are recorded in the specified variable, and the error causes are analyzed and the correction methods are proposed; if no errors are found, "no error" is recorded.

[0043] For example, the subtask instruction of result verification is: checking the calculation or reasoning process, confirming whether the equation application, data reference is correct, and the information extraction is complete. If problems are found, the error conditions are recorded in the variable "{error_detail}", including error type (such as formula error, data extraction omission, etc.), specific steps and position of error occurrence; at the same time, the correction method for the error is recorded in the variable "{correction_method}", such as correcting the formula, supplementing the missing data, etc. If there is no problem, record "no error" (corresponding to the error condition) and "none" (corresponding to the correction method).

[0044] At step 250, according to the sub-task instruction of answer analysis, the question and answer model optimizes and analyzes the preliminary answer according to the verification result, and outputs the question and answer result.

[0045] Specifically, according to the requirement of the sub-task instruction of answer analysis, the question and answer model optimizes the preliminary answer according to the verification result, determines the final answer, and generates answer analysis according to the final answer.

[0046] From the actual application scene and user demand, the question and answer model optimizes and analyzes the preliminary answer according to the verification result. For example, if the answer is a numerical value, reasonable rounding or decimal retention is performed according to the actual meaning, and the unit is specified; if the answer is text information, language polishing is performed to make it clearer and easier to understand. The final answer and the answer analysis are stored in the specified variables, respectively, to provide complete and valuable answers for users.

[0047] For example, the sub-task instruction of answer analysis is: according to the verification result (if there is an error, the correction method recorded is required to be modified), determine the final answer, and store the final answer in the variable "{final_answer}". And from the perspective of rail transit operation, passenger travel, etc. The answer is briefly analyzed to generate answer analysis, and the answer analysis is stored in the variable "{answer_explanation}".

[0048] The question and answer method of the rail transit table-text mixed data provided by the application realizes efficient and accurate question and answer of complex table-text mixed data in the field of rail transit through the chain process of information extraction, logical construction, preliminary operation, result verification and answer analysis. The output of each link can be traced back, further enhancing the explainability of the question and answer; through the design of sub-task instructions in stages, efficient and accurate question and answer of complex table-text mixed data in the field of rail transit is realized.

[0049] Compared with the traditional end-to-end method, the cognitive chain reasoning framework realizes fine control and optimization of the reasoning process by generating a series of intermediate results. Each part focuses on a specific reasoning task, making it possible to optimize the performance of each link, thereby significantly improving the overall efficiency and accuracy of the framework in processing complex discrete reasoning tasks. It provides an advanced and effective solution to the question and answer problem of integrating table and text data.

[0050] In some embodiments, the question and answer result includes key information, reasoning logic, preliminary answer, verification result, final answer and answer analysis.

[0051] Specifically, in order to further improve the explainability of the question and answer, a series of intermediate results in the reasoning process are also output, therefore, the question and answer result includes not only the final answer and the answer analysis, but also the key information, the reasoning logic, the preliminary answer and the verification result.

[0052] In some embodiments, an output format sub-task instruction can be added in the task instruction to require the question and answer model to output key information, reasoning logic, preliminary answer, verification result, final answer and answer analysis.

[0053] For example, the output format sub-task instruction is: arrange the results in the following format: step 1 outputs {evidence}, step 2 outputs {equation_or_logic}, step 3 outputs {preliminary_answer}, step 4 outputs {error_detail} #{correction_method}, step 5 outputs {final_answer} #{answer_explanation}, the format of the final answer is: the answer is: {final_answer}, and the answer analysis is: {answer_explanation}.

[0054] In some embodiments, the question and answer method for rail transit table-text mixed data provided by the present application comprises: constructing a training sample containing task instructions, tables, texts, questions and answers; pre-training the question and answer model based on the training sample.

[0055] Specifically, existing large language models are difficult to internalize the calculation rules specific to rail transit (such as passenger flow calculation formula, train dispatching priority rule), often resulting in unit conversion errors, misinterpretation of professional terms, and poor adherence to domain rules.

[0056] To solve the above problems, a training sample containing task instructions, tables, texts, questions and answers is constructed, and the question and answer model is pre-trained based on the five-tuple training sample to accurately depict the domain knowledge association and reasoning path dependence.

[0057] The task instruction adopts a task decomposition strategy, decomposing complex reasoning into executable atomic operations (information extraction, logic construction, preliminary operation, result verification and answer analysis); tables and texts provide multi-element real data and detailed descriptions, providing rich basis for reasoning; questions are carefully set around tables and texts, including numerical calculation, logical judgment, information extraction, etc., to comprehensively test the reasoning ability of the model in complex discrete situations; the answer content not only gives the final answer, but also records the intermediate results of each reasoning step, facilitating model review learning, and further optimizing the reasoning process and answer accuracy.

[0058] The question and answer method for rail transit table-text mixed data provided by the present application enhances the rule adherence of the question and answer model by constructing a training sample containing task instructions, tables, texts, questions and answers, and pre-training the question and answer model based on the training sample.

[0059] In some embodiments, the application provides a question and answer method for rail transit table-text hybrid data, which further comprises: Fine-tuning the pre-trained question and answer model by using a low-rank adaptive matrix.

[0060] Specifically, to solve the problem of high transfer cost in the field of general large models, the pre-trained question and answer model is fine-tuned by using a low-rank adaptive matrix, that is, the weights of the pre-trained model are adapted by introducing a low-rank adaptive matrix (low-dimensional trainable parameters), the original parameters are frozen, and only the newly added lightweight module is trained.

[0061] Through the low-rank decomposition strategy of the parameter matrix, the adjustable parameter scale is compressed to 8.3% of the original model under the premise of preserving the integrity of the pre-trained knowledge, so that the general large model quickly adapts to the table parsing, multi-step calculation and logic verification requirements of the rail transit scene, forming a table-text hybrid data question and answer model dedicated to the rail transit field.

[0062] The question and answer method for rail transit table-text hybrid data provided by the application fine-tunes the pre-trained question and answer model by using a low-rank adaptive matrix, which reduces the transfer cost of general large models. Compared with traditional architecture, this fine-tuning scheme not only significantly improves the model convergence speed (reduces the number of training iterations by 65%), but also effectively controls the consumption of computing resources.

[0063] In actual application scene tests, the question and answer method provided by the application has a joint parsing accuracy of 92.7% for structured tables and unstructured texts, especially in multi-step logical deduction tasks, it can effectively establish an information association network and realize deep coupling of data features.

[0064] The question and answer device for rail transit table-text hybrid data provided by the application is described below, and the question and answer device for rail transit table-text hybrid data described below can be mutually corresponding and referred to the question and answer method for rail transit table-text hybrid data described above.

[0065] Figure 3 is a structural schematic diagram of the question and answer device for rail transit table-text hybrid data provided by the application, as Figure 3 shown, the application provides a question and answer device for rail transit table-text hybrid data, which comprises: A cognitive chain reasoning module 310 is configured to input the rail transit table-text hybrid data and the question into a question and answer model, the question and answer model performs reasoning according to the task instruction, and outputs the question and answer result. The task instruction is constructed based on a cognitive chain inference framework including five steps of information extraction, logic construction, preliminary operation, result verification, and answer analysis, and the task instruction includes five sub-task instruction of information extraction, logic construction, preliminary operation, result verification, and answer analysis.

[0066] In some embodiments, the cognitive chain inference module 310 includes: an information extraction unit configured to extract, according to the sub-task instruction of information extraction, key information related to the question from the rail transit table-text hybrid data by the question and answer model; a logic construction unit configured to construct, according to the sub-task instruction of logic construction, inference logic based on the key information and professional knowledge and conventional rules in the rail transit field by the question and answer model; a preliminary operation unit configured to operate, according to the sub-task instruction of preliminary operation, the key information according to the inference logic to obtain a preliminary answer by the question and answer model; a result verification unit configured to verify the preliminary answer to obtain a verification result by the question and answer model according to the sub-task instruction of result verification; an answer analysis unit configured to optimize and analyze the preliminary answer according to the verification result to output the question and answer result by the question and answer model according to the sub-task instruction of answer analysis.

[0067] In some embodiments, the device further includes a pre-training module configured to: construct a training sample including the task instruction, table, text, question, and answer; pre-train the question and answer model based on the training sample.

[0068] In some embodiments, the device further includes a fine-tuning module configured to: fine-tune the pre-trained question and answer model by using a low-rank adaptive matrix.

[0069] In some embodiments, the question and answer result includes the key information, the inference logic, the preliminary answer, the verification result, the final answer, and the answer analysis.

[0070] It should be noted that the above question and answer device for rail transit table-text hybrid data provided by the present application can realize all the method steps realized by the above method embodiments and achieve the same technical effects. Therefore, the same parts and beneficial effects of the method embodiments will not be described in detail.

[0071] Figure 4 is a structural schematic diagram of an electronic device provided by the present application, asFigure 4 As shown, the electronic device can include a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communications bus 440. The processor 410 can invoke the logical instructions in the memory 430 to perform the question and answer method of rail transit table-text mixed data, which includes inputting the rail transit table-text mixed data and a question into a question and answer model, wherein the question and answer model performs reasoning according to a task instruction and outputs a question and answer result; wherein the task instruction is constructed based on a cognitive chain reasoning framework including five steps of information extraction, logic construction, preliminary operation, result verification, and answer analysis, and the task instruction includes five sub-task instructions of information extraction, logic construction, preliminary operation, result verification, and answer analysis.

[0072] In addition, the logical instructions in the memory 430 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0073] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the question and answer method of rail transit table-text mixed data provided by the above-mentioned methods, which includes inputting the rail transit table-text mixed data and a question into a question and answer model, wherein the question and answer model performs reasoning according to a task instruction and outputs a question and answer result; wherein the task instruction is constructed based on a cognitive chain reasoning framework including five steps of information extraction, logic construction, preliminary operation, result verification, and answer analysis, and the task instruction includes five sub-task instructions of information extraction, logic construction, preliminary operation, result verification, and answer analysis.

[0074] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the question-answering method of the rail transit table-text hybrid data provided by the above method, the method comprising: inputting the rail transit table-text hybrid data and a question into a question-answering model, the question-answering model performing inference according to a task instruction, and outputting a question-answering result; wherein the task instruction is constructed based on a cognitive chain inference framework comprising five steps of information extraction, logic construction, preliminary operation, result verification, and answer analysis, and the task instruction comprises five sub-task instructions of information extraction, logic construction, preliminary operation, result verification, and answer analysis.

[0075] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0076] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software plus necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0077] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A question and answer method of rail transit table-text hybrid data, characterized in that, The method comprises the following steps: inputting the rail transit table-text mixed data and a question into a question-answering model, and outputting a question-answering result by reasoning according to a task instruction; wherein the task instruction is constructed based on a cognitive chain reasoning framework comprising five steps of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the task instruction comprises five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis.

2. The rail transit table-text hybrid data question and answer method according to claim 1, characterized in that, The method further comprises the following steps: constructing a training sample comprising the task instruction, table, text, question and answer; pre-training the question-answering model based on the training sample. The method further comprises the following steps: fine-tuning the pre-trained question-answering model by using a low-rank adaptive matrix. The question-answering result comprises the key information, reasoning logic, preliminary answer, verification result, final answer and answer analysis. 3.The rail transit table-text hybrid data question and answer method according to claim 1, characterized in that, The question-answering model is a DeepSeek R1 large model. The method comprises the following steps: a cognitive chain reasoning module is configured to input rail transit table-text mixed data and a question into a question-answering model, and output a question-answering result by reasoning according to a task instruction; 4. The rail transit table-text hybrid data question and answer method according to claim 3, characterized in that, wherein the task instruction is constructed based on a cognitive chain reasoning framework comprising five steps of information extraction, logic construction, preliminary operation, result verification and answer analysis, and the task instruction comprises five sub-task instructions of information extraction, logic construction, preliminary operation, result verification and answer analysis. The processor executes the computer program to implement the question-answering method for the rail transit table-text mixed data according to any one of claims 1 to 6.

5. The rail transit table-text hybrid data question and answer method according to claim 2, characterized in that, The computer program is executed by the processor to implement the question-answering method for the rail transit table-text mixed data according to any one of claims 1 to 6.

6. The rail transit table-text hybrid data question and answer method according to any one of claims 1-5, characterized in that, The computer program is executed by the processor to implement the question-answering method for the rail transit table-text mixed data according to any one of claims 1 to 6.

7. A question and answer device for rail transit table-text hybrid data, characterized in that, ​ ​ ​ 8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, ​ 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, ​ 10. A computer program product comprising a computer program, characterized in that, ​

Images